Midjourney Prompt Guide 2026: Parameters & v7 Syntax

This midjourney prompt guide covers Midjourney v7 parameters and trailing syntax as of mid-2026. You write a subject-first scene line, then append flags: --ar, --stylize (or --s), --style raw or --raw, --chaos, --quality, --no, --v 7, plus reference controls --sref, --oref, and related weights. Wrong spacing, buried flags, or a missing version turn a solid brief into wasted credits. You leave with a parameter map, placement rules, reference syntax, a short workflow, and common mistakes. Soft path when you start from a photo: PromptMake /image. Sister posts cover daily prompt examples and single-dial deep dives; this page owns the control strip.

Who this midjourney prompt guide is for

You run Midjourney v7 on the web app or Discord and need the flags to behave. Art directors share strings across a team and need one syntax standard. Product designers switch between square catalog and 9:16 social and keep forgetting which --ar goes with which brief. Illustrators raise stylize for concept art and forget to drop it when the next job is a packshot. Anyone pasting FLUX-style essays into Midjourney without a parameter stack lands here too.

Skip this page if you only want copy-paste scene examples. The examples-led Midjourney playbook on this site covers daily structure and sample prompts. Stay here when the question is syntax: where flags go, what each dial does, how --sref and --oref attach, and which mistakes burn Fast hours. V8 Alpha may appear on some accounts; keep --v 7 explicit when you compare or hand off strings so version drift does not fake a quality drop.

Clear likeness and brand rights before you publish reference-driven stills. Treat generator grids as comps until legal signs off.

Midjourney v7 prompt anatomy

A Midjourney v7 prompt has two zones. Zone one is scene language: subject, action, place, light, medium. Zone two is the parameter strip: double-dash flags after the prose. Midjourney still weights early nouns heavily, so put the hero first. Parameters always trail. Do not bury --ar 16:9 between clauses; the bot may treat fragments as text or ignore the flag.

v7 reads natural sentences better than old tag soup. You still win with short clauses and one medium word: editorial photography, watercolor concept art, product catalog shot. Mix "whimsical storybook" with catalog photo language in one string and the model splits the difference. Pick a register, then tune the strip. Photographic jobs lean on --style raw (or --raw where your build accepts it) plus low-to-mid stylize. Painterly jobs skip raw and raise stylize so the house aesthetic helps.

Official docs list many flags. Daily production uses a short core: aspect, stylize, raw mode, chaos, quality, version, --no, and the reference family. Personalization (--profile / --p), --tile, --stop, and seed tools matter for niche jobs. Learn the core strip first. Add niche flags when a brief demands them. The subsections below fix syntax rules, then walk the dials you change every week.

Syntax rules that prevent silent failures

Use a space before each --. Write --ar 16:9, never - -ar or --ar16:9. Do not put commas between parameters. Do not place scene words after the parameter block; text after flags can fail or confuse the parse. One prompt line, parameters at the end, space-separated flags.

Short forms work where Midjourney documents them: --ar for --aspect, --s for --stylize, --c for --chaos, --q for --quality, --p for --profile. Prefer the long form in team templates so juniors read the intent. Keep --v 7 on shared strings even if your account defaults to v7. Defaults change; explicit version keeps archives honest.

Scene line shape before the flags

Default shape: [Subject + action] + [place] + [light] + [camera or medium] + [parameters]. Keep mood to one short clause. Name materials when speculars matter. For photoreal work, camera language (85mm, soft window light, shallow depth) pairs with raw and modest stylize. For illustration, medium words (gouache, ink wash, concept matte painting) pair with higher stylize and no raw.

Multi-prompt weights (::) still exist for advanced splits. Most commercial jobs never need them. Fix subject, light, and one parameter first. Reach for weights when two concepts fight inside one frame after the strip is already clean.

Core Midjourney v7 parameters

Treat the parameter strip as a control panel, not decoration. Each flag answers a production question: frame shape, how hard house taste pushes, how literal the camera language stays, how wild the four-up grid spreads, how much detail spend you buy, and which model build runs. Change one dial per iteration when you debug. Changing stylize, chaos, and raw together hides the cause of a bad grid.

As of mid-2026, Midjourney’s public parameter list still centers on --aspect / --ar, --chaos / --c, --stylize / --s, Raw Mode, --quality / --q, --no, Style Reference, Omni Reference (identity and object lock on v7), Personalization, --tile, and related tools. Exact UI labels shift; the trailing syntax below matches common Discord and web Imagine usage. Confirm any brand-new flag on docs.midjourney.com before you bake it into a client SOP.

Aspect, version, quality, chaos

--ar W:H sets frame shape inside Midjourney. Common jobs: --ar 1:1 catalog, --ar 4:5 or --ar 2:3 portrait, --ar 16:9 landscape or hero, --ar 9:16 stories. Match deliverable before you chase style. Upscales inherit the ratio you chose.

--v 7 selects Midjourney v7. Pin it on templates. --quality / --q trades detail and GPU time; higher quality costs more and takes longer. Start at default for exploration; raise quality on finals you already like. --chaos / --c (0–100) spreads variety across the four-image grid. Low chaos (0–15) keeps tight variants for client reviews. Higher chaos explores composition when the brief is loose.

Stylize, raw mode, and --no

--stylize / --s runs 0–1000 (default often 100). Low values stay literal. High values push Midjourney’s house aesthetic. Photoreal with raw often sits near 50–150. Concept illustration often sits near 200–500. Deep stylize dives live in the dedicated stylize article; here you only need the dial’s job on the strip.

--style raw or --raw (builds differ on which form they surface) reduces automatic beautification. Use it when photographic words keep turning painterly. Skip it when you want Midjourney polish for fantasy or editorial illustration. Pair raw with lower stylize; high stylize fights raw.

--no excludes concepts: --no text, watermark, logo or --no crowd. Keep the list short. Long exclusion lists fight the scene nouns. For empty seamless product work, name the empty set in prose and add a tight --no for clutter if needed.

Reference parameters: --sref, --oref, and weights

Reference flags do work prose cannot. Style Reference (--sref) pulls palette, grade, and finish from an image URL or a style code. Omni Reference (--oref) locks a person’s likeness or an object’s form on v7 (it covers jobs older Character Reference flows handled). Style weight (--sw) and omni/character weight dials (--ow / --cw where your build exposes them) set how hard the reference pushes. You change scene text while the reference strip stays fixed across a campaign.

Combine carefully. --sref owns look. --oref owns identity or object geometry. Tune one weight at a time. Max weight on both can freeze the frame into a near-copy of the references and ignore new scene language. Start mid-range, generate once, then nudge. Style Reference versions (--sv) exist for how Midjourney interprets an --sref; legacy codes sometimes need an older --sv value. Check current docs when an old code looks wrong on v7.

Image prompts (URLs at the start of the prompt) still bias composition toward the reference photo. They are not the same as --sref. Image prompts lean content and layout. --sref leans aesthetic. Use image prompts when the composition must echo a still; use --sref when the grade and finish must match a lookbook plate. Soft path if you hold a photo and need a Midjourney-shaped draft before you attach references: https://promptmake.net/image

--sref syntax patterns

Trailing form: scene text --sref https://example.com/grade.jpg --sw 100 --v 7. Multiple URLs space-separated after --sref. Numeric style codes use --sref CODE where Midjourney’s library supports them. Pair with --ar and raw/stylize as usual. Keep --sw moderate until the look sticks; over-weighting can wash out subject detail.

Save approved --sref URLs in a team sheet with date and --sv notes. Codes and algorithms change. A working campaign strip from early 2025 may need an --sv pin or a fresh reference still by mid-2026.

--oref and identity lock

Omni Reference attaches after the scene: scene text --oref https://example.com/face-or-object.jpg plus the weight flag your build documents. Use a clean, well-lit reference. Busy backgrounds leak into the lock. For product geometry, use a clear packshot as the omni source and keep --style raw on so house polish does not reshape the silhouette.

Do not rely on celebrity likeness hacks or scraped faces. Use owned talent, staff with consent, or brand mascots you control. Reference strength is a production tool, not a shortcut around rights.

Step-by-step parameter workflow

Ship a repeatable strip instead of reinventing flags each morning. Lock aspect and version first. Write the scene. Add raw and stylize for the medium. Generate. Score subject, light, and frame. Change one parameter. Only then attach --sref or --oref when consistency across a set matters. Teams that reverse this order spend hours blaming the model for a wrong --ar or an accidental high chaos default.

Keep a one-line template per job type in a shared note: catalog square, portrait 4:5, landscape 16:9, story 9:16. Each template holds --ar, default stylize band, raw on/off, chaos ceiling, and --v 7. Paste the template, swap scene nouns, run. After approval, clone the strip onto the next SKU or pose. The subsections walk first grid, iteration, and photo-assisted drafting.

Budget one Fast hour the first week you standardize strips. Later briefs take minutes when the templates already match your deliverables. Personalization profiles (--p) sit on top for accounts that trained a custom look; treat them like a soft --sref and document when the profile is on.

1. Lock frame and version

Write --ar for the final crop and --v 7 before you polish adjectives. Wrong ratio forces a re-render after art direction already passed. Confirm the account default version matches the string you archive.

2. Set raw and stylize for the medium

Photographic brief: add --style raw (or --raw) and --stylize in the 50–150 band. Illustrated brief: skip raw, raise stylize into the 200–500 band, keep chaos modest if the client needs similar options. Product geometry: raw on, stylize low, chaos low.

3. Iterate one dial, then add references

If the grid is too samey, raise chaos slightly. If it ignores the brief, lower stylize. If it looks over-painted, confirm raw is present. When subject and light pass, attach --sref for grade and --oref for identity. Soft draft from a still: upload to PromptMake /image, choose Midjourney as the write-out, pick Recreate Exactly, Change Style, Adjust Lighting, or Create Variation, then paste your parameter strip on the draft. Guests get about 3 image generations per day; free registration raises the image quota. Image and text quotas stay separate: https://promptmake.net/image

Common parameter mistakes

Mistake 1: Flags in the middle of the sentence. Move every -- token to the end.

Mistake 2: Spaces inside the double dash (- -ar). Midjourney misses the flag.

Mistake 3: Commas between parameters. Use spaces only.

Mistake 4: High stylize with raw on a product brief. The dials fight; drop stylize.

Mistake 5: Forgetting --v 7 on shared prompts while someone else still defaults to an older model.

Mistake 6: Maxing --sw and omni weight until the scene text stops mattering.

Mistake 7: Pasting Midjourney flags into FLUX or GPT Image fields. Strip the control strip before cross-host paste.

Mistake 8: Using --no as a substitute for positive empty-scene language. Name the clean set, then exclude only stubborn artifacts.

Mistake 9: Changing five flags after one bad grid. Isolate the failure layer.

Mistake 10: Treating this midjourney prompt guide as a substitute for rights review on --oref faces and brand packs.

Soft next step on PromptMake /image

PromptMake /image fits when you hold a reference photo and need Midjourney-shaped prose before you attach your parameter strip. Upload a still, pick a goal mode, select Midjourney, copy the draft, then append --ar, raw, stylize, --v 7, and any --sref / --oref your campaign uses. Soft start: https://promptmake.net/image

Blank-page Midjourney example libraries and FLUX dialect comparisons live elsewhere. Stylize-only and --sref-only deep dives live elsewhere. This midjourney prompt guide owns parameters and v7 syntax so your control strip stays consistent while scene nouns change.

FAQ

What does a midjourney prompt guide cover in 2026?

A midjourney prompt guide for 2026 should explain Midjourney v7 scene shape plus the trailing parameter strip: aspect, stylize, raw mode, chaos, quality, version, --no, and reference flags. Structure articles teach what to write. Parameter guides teach how flags steer the same words. You need both when teams share prompts across Discord and the web app.

Where do Midjourney v7 parameters go in the prompt?

Put every parameter at the end of the prompt, space-separated, with a space before each --. Do not insert flags between scene clauses. Do not put prose after the parameter block. Correct form: scene text, then --ar 4:5 --style raw --stylize 100 --v 7. Wrong placement is a common silent failure.

What is the difference between --stylize and --style raw?

--stylize scales how hard Midjourney applies house aesthetic taste (0–1000). --style raw / --raw reduces automatic beautification so camera language reads more literally. Use both on photoreal jobs: raw on, stylize low-to-mid. Raise stylize and skip raw for painterly concept work. They are separate dials, not synonyms.

How do --sref and --oref work on Midjourney v7?

--sref locks style, palette, and grade from a URL or style code. --oref locks a person or object form (Omni Reference on v7). Tune strength with --sw and the omni/character weight your build documents. Combine them for campaign consistency, but start at moderate weights so scene text still drives pose and setting.

Which Midjourney parameters should I set first?

Set --ar and --v 7 first so the frame and model match the deliverable. Add raw and stylize for the medium next. Adjust chaos when the four-up grid is too tight or too wild. Attach --sref and --oref after subject and light already pass. Quality bumps belong on near-final picks, not on every exploratory run.

Can I reuse the same parameter strip across prompts?

Yes. Save strips per job type (catalog 1:1, portrait 4:5, hero 16:9) with stylize band, raw on/off, chaos ceiling, and --v 7. Swap only the scene nouns. Add campaign --sref URLs once art direction approves a grade plate. Reusing strips cuts version drift and speeds handoffs.

How do I start if I only have a reference photo?

Upload the photo to PromptMake /image, select Midjourney as the output dialect, choose Recreate Exactly or Create Variation, copy the draft, then append your parameter strip. Soft start: https://promptmake.net/image. Guests get about 3 image generations per day; free registration raises the image quota. Confirm rights on any face or pack before you publish.